xiaoou
f43e47b0a6
feat: V-learning架构 — 改Q→V
Q-learning问题: pass action Q值稳定偏高 → 永远pass
V-learning修复:
- 网络输出 14(Q-action)→1(V-state): P(win|state)
- NeuralBot: 对每个候选编码出牌后state → 选V值最高的
- pass和出牌都是候选, 公平竞争(不是固定action slot)
TrainingCollector:
- 新增 EncodeAfterPlay(view, tracker, afterHand, playedIds)
- EncodeState 支持 extraSeen 参数(模拟出牌时标记已见)
Python:
- train_v.py: V-learning训练器(load CSV→predict win/loss)
- network.py: OUTPUT_DIM=14→1
验证: 200局0错误, V-model loss 0.81→0.0001
需更多数据覆盖状态空间
2026-07-13 02:52:48 +08:00
..
2026-07-13 01:05:14 +08:00
2026-07-13 00:40:27 +08:00
2026-07-12 22:26:29 +08:00
2026-07-13 02:52:48 +08:00
2026-07-12 22:40:49 +08:00
2026-07-13 02:52:48 +08:00
2026-07-07 14:39:50 +08:00
2026-07-07 12:02:15 +08:00
2026-07-07 12:02:15 +08:00
2026-07-13 02:52:48 +08:00
2026-07-13 00:40:27 +08:00
2026-07-13 02:52:48 +08:00